Add GGUF README
Browse files- GGUF/README.md +31 -0
GGUF/README.md
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# GGUF Models
|
| 2 |
+
|
| 3 |
+
This folder contains quantized GGUF versions of the CardVault+ model for efficient inference with llama.cpp.
|
| 4 |
+
|
| 5 |
+
## Available Models
|
| 6 |
+
|
| 7 |
+
| Model File | Size | Quantization | Use Case |
|
| 8 |
+
|---|---|---|---|
|
| 9 |
+
| `cardvault-500m-f16.gguf` | 783MB | F16 (Base) | Maximum quality |
|
| 10 |
+
| `cardvault-500m-mmproj-f16.gguf` | 191MB | F16 (Vision) | **REQUIRED** |
|
| 11 |
+
| `cardvault-500m-q8_0.gguf` | 417MB | Q8_0 | Near-perfect quality |
|
| 12 |
+
| `cardvault-500m-q6_k.gguf` | 399MB | Q6_K | Balanced |
|
| 13 |
+
| `cardvault-500m-q5_k_m.gguf` | 311MB | Q5_K_M | **Recommended** |
|
| 14 |
+
| `cardvault-500m-q4_k_m.gguf` | 290MB | Q4_K_M | Maximum compression |
|
| 15 |
+
|
| 16 |
+
## Usage
|
| 17 |
+
|
| 18 |
+
```bash
|
| 19 |
+
# Download llama.cpp
|
| 20 |
+
git clone https://github.com/ggerganov/llama.cpp
|
| 21 |
+
cd llama.cpp && make
|
| 22 |
+
|
| 23 |
+
# Run inference (Q5_K_M recommended)
|
| 24 |
+
./main \
|
| 25 |
+
--model cardvault-500m-q5_k_m.gguf \
|
| 26 |
+
--mmproj cardvault-500m-mmproj-f16.gguf \
|
| 27 |
+
--image credit_card.jpg \
|
| 28 |
+
--prompt "Extract card information in JSON format"
|
| 29 |
+
```
|
| 30 |
+
|
| 31 |
+
⚠️ **Critical**: Both text model + mmproj required for vision functionality!
|